Commit | Line | Data |
---|---|---|
5274f052 JA |
1 | /* |
2 | * "splice": joining two ropes together by interweaving their strands. | |
3 | * | |
4 | * This is the "extended pipe" functionality, where a pipe is used as | |
5 | * an arbitrary in-memory buffer. Think of a pipe as a small kernel | |
6 | * buffer that you can use to transfer data from one end to the other. | |
7 | * | |
8 | * The traditional unix read/write is extended with a "splice()" operation | |
9 | * that transfers data buffers to or from a pipe buffer. | |
10 | * | |
11 | * Named by Larry McVoy, original implementation from Linus, extended by | |
c2058e06 JA |
12 | * Jens to support splicing to files, network, direct splicing, etc and |
13 | * fixing lots of bugs. | |
5274f052 | 14 | * |
c2058e06 JA |
15 | * Copyright (C) 2005-2006 Jens Axboe <axboe@suse.de> |
16 | * Copyright (C) 2005-2006 Linus Torvalds <torvalds@osdl.org> | |
17 | * Copyright (C) 2006 Ingo Molnar <mingo@elte.hu> | |
5274f052 JA |
18 | * |
19 | */ | |
20 | #include <linux/fs.h> | |
21 | #include <linux/file.h> | |
22 | #include <linux/pagemap.h> | |
23 | #include <linux/pipe_fs_i.h> | |
24 | #include <linux/mm_inline.h> | |
5abc97aa | 25 | #include <linux/swap.h> |
4f6f0bd2 JA |
26 | #include <linux/writeback.h> |
27 | #include <linux/buffer_head.h> | |
a0f06780 | 28 | #include <linux/module.h> |
4f6f0bd2 | 29 | #include <linux/syscalls.h> |
912d35f8 | 30 | #include <linux/uio.h> |
5274f052 | 31 | |
912d35f8 JA |
32 | struct partial_page { |
33 | unsigned int offset; | |
34 | unsigned int len; | |
35 | }; | |
36 | ||
37 | /* | |
00522fb4 | 38 | * Passed to splice_to_pipe |
912d35f8 JA |
39 | */ |
40 | struct splice_pipe_desc { | |
41 | struct page **pages; /* page map */ | |
42 | struct partial_page *partial; /* pages[] may not be contig */ | |
43 | int nr_pages; /* number of pages in map */ | |
44 | unsigned int flags; /* splice flags */ | |
45 | struct pipe_buf_operations *ops;/* ops associated with output pipe */ | |
46 | }; | |
47 | ||
83f9135b JA |
48 | /* |
49 | * Attempt to steal a page from a pipe buffer. This should perhaps go into | |
50 | * a vm helper function, it's already simplified quite a bit by the | |
51 | * addition of remove_mapping(). If success is returned, the caller may | |
52 | * attempt to reuse this page for another destination. | |
53 | */ | |
5abc97aa JA |
54 | static int page_cache_pipe_buf_steal(struct pipe_inode_info *info, |
55 | struct pipe_buffer *buf) | |
56 | { | |
57 | struct page *page = buf->page; | |
4f6f0bd2 | 58 | struct address_space *mapping = page_mapping(page); |
5abc97aa | 59 | |
9e0267c2 JA |
60 | lock_page(page); |
61 | ||
5abc97aa JA |
62 | WARN_ON(!PageUptodate(page)); |
63 | ||
ad8d6f0a JA |
64 | /* |
65 | * At least for ext2 with nobh option, we need to wait on writeback | |
66 | * completing on this page, since we'll remove it from the pagecache. | |
67 | * Otherwise truncate wont wait on the page, allowing the disk | |
68 | * blocks to be reused by someone else before we actually wrote our | |
69 | * data to them. fs corruption ensues. | |
70 | */ | |
71 | wait_on_page_writeback(page); | |
72 | ||
4f6f0bd2 JA |
73 | if (PagePrivate(page)) |
74 | try_to_release_page(page, mapping_gfp_mask(mapping)); | |
75 | ||
9e0267c2 JA |
76 | if (!remove_mapping(mapping, page)) { |
77 | unlock_page(page); | |
5abc97aa | 78 | return 1; |
9e0267c2 | 79 | } |
5abc97aa | 80 | |
0568b409 | 81 | buf->flags |= PIPE_BUF_FLAG_LRU; |
5abc97aa JA |
82 | return 0; |
83 | } | |
84 | ||
5274f052 JA |
85 | static void page_cache_pipe_buf_release(struct pipe_inode_info *info, |
86 | struct pipe_buffer *buf) | |
87 | { | |
88 | page_cache_release(buf->page); | |
89 | buf->page = NULL; | |
0568b409 | 90 | buf->flags &= ~PIPE_BUF_FLAG_LRU; |
5274f052 JA |
91 | } |
92 | ||
93 | static void *page_cache_pipe_buf_map(struct file *file, | |
94 | struct pipe_inode_info *info, | |
95 | struct pipe_buffer *buf) | |
96 | { | |
97 | struct page *page = buf->page; | |
49d0b21b | 98 | int err; |
5274f052 JA |
99 | |
100 | if (!PageUptodate(page)) { | |
49d0b21b JA |
101 | lock_page(page); |
102 | ||
103 | /* | |
104 | * Page got truncated/unhashed. This will cause a 0-byte | |
73d62d83 | 105 | * splice, if this is the first page. |
49d0b21b JA |
106 | */ |
107 | if (!page->mapping) { | |
108 | err = -ENODATA; | |
109 | goto error; | |
110 | } | |
5274f052 | 111 | |
49d0b21b | 112 | /* |
73d62d83 | 113 | * Uh oh, read-error from disk. |
49d0b21b JA |
114 | */ |
115 | if (!PageUptodate(page)) { | |
116 | err = -EIO; | |
117 | goto error; | |
118 | } | |
119 | ||
120 | /* | |
73d62d83 | 121 | * Page is ok afterall, fall through to mapping. |
49d0b21b | 122 | */ |
5274f052 | 123 | unlock_page(page); |
5274f052 JA |
124 | } |
125 | ||
49d0b21b JA |
126 | return kmap(page); |
127 | error: | |
128 | unlock_page(page); | |
129 | return ERR_PTR(err); | |
5274f052 JA |
130 | } |
131 | ||
132 | static void page_cache_pipe_buf_unmap(struct pipe_inode_info *info, | |
133 | struct pipe_buffer *buf) | |
134 | { | |
5274f052 JA |
135 | kunmap(buf->page); |
136 | } | |
137 | ||
912d35f8 JA |
138 | static void *user_page_pipe_buf_map(struct file *file, |
139 | struct pipe_inode_info *pipe, | |
140 | struct pipe_buffer *buf) | |
141 | { | |
142 | return kmap(buf->page); | |
143 | } | |
144 | ||
145 | static void user_page_pipe_buf_unmap(struct pipe_inode_info *pipe, | |
146 | struct pipe_buffer *buf) | |
147 | { | |
148 | kunmap(buf->page); | |
149 | } | |
150 | ||
70524490 JA |
151 | static void page_cache_pipe_buf_get(struct pipe_inode_info *info, |
152 | struct pipe_buffer *buf) | |
153 | { | |
154 | page_cache_get(buf->page); | |
155 | } | |
156 | ||
5274f052 JA |
157 | static struct pipe_buf_operations page_cache_pipe_buf_ops = { |
158 | .can_merge = 0, | |
159 | .map = page_cache_pipe_buf_map, | |
160 | .unmap = page_cache_pipe_buf_unmap, | |
161 | .release = page_cache_pipe_buf_release, | |
5abc97aa | 162 | .steal = page_cache_pipe_buf_steal, |
70524490 | 163 | .get = page_cache_pipe_buf_get, |
5274f052 JA |
164 | }; |
165 | ||
912d35f8 JA |
166 | static int user_page_pipe_buf_steal(struct pipe_inode_info *pipe, |
167 | struct pipe_buffer *buf) | |
168 | { | |
169 | return 1; | |
170 | } | |
171 | ||
172 | static struct pipe_buf_operations user_page_pipe_buf_ops = { | |
173 | .can_merge = 0, | |
174 | .map = user_page_pipe_buf_map, | |
175 | .unmap = user_page_pipe_buf_unmap, | |
176 | .release = page_cache_pipe_buf_release, | |
177 | .steal = user_page_pipe_buf_steal, | |
178 | .get = page_cache_pipe_buf_get, | |
179 | }; | |
180 | ||
83f9135b JA |
181 | /* |
182 | * Pipe output worker. This sets up our pipe format with the page cache | |
183 | * pipe buffer operations. Otherwise very similar to the regular pipe_writev(). | |
184 | */ | |
00522fb4 JA |
185 | static ssize_t splice_to_pipe(struct pipe_inode_info *pipe, |
186 | struct splice_pipe_desc *spd) | |
5274f052 | 187 | { |
912d35f8 | 188 | int ret, do_wakeup, page_nr; |
5274f052 JA |
189 | |
190 | ret = 0; | |
191 | do_wakeup = 0; | |
912d35f8 | 192 | page_nr = 0; |
5274f052 | 193 | |
3a326a2c IM |
194 | if (pipe->inode) |
195 | mutex_lock(&pipe->inode->i_mutex); | |
5274f052 | 196 | |
5274f052 | 197 | for (;;) { |
3a326a2c | 198 | if (!pipe->readers) { |
5274f052 JA |
199 | send_sig(SIGPIPE, current, 0); |
200 | if (!ret) | |
201 | ret = -EPIPE; | |
202 | break; | |
203 | } | |
204 | ||
6f767b04 JA |
205 | if (pipe->nrbufs < PIPE_BUFFERS) { |
206 | int newbuf = (pipe->curbuf + pipe->nrbufs) & (PIPE_BUFFERS - 1); | |
3a326a2c | 207 | struct pipe_buffer *buf = pipe->bufs + newbuf; |
5274f052 | 208 | |
912d35f8 JA |
209 | buf->page = spd->pages[page_nr]; |
210 | buf->offset = spd->partial[page_nr].offset; | |
211 | buf->len = spd->partial[page_nr].len; | |
212 | buf->ops = spd->ops; | |
6f767b04 | 213 | pipe->nrbufs++; |
912d35f8 JA |
214 | page_nr++; |
215 | ret += buf->len; | |
216 | ||
6f767b04 JA |
217 | if (pipe->inode) |
218 | do_wakeup = 1; | |
5274f052 | 219 | |
912d35f8 | 220 | if (!--spd->nr_pages) |
5274f052 | 221 | break; |
6f767b04 | 222 | if (pipe->nrbufs < PIPE_BUFFERS) |
5274f052 JA |
223 | continue; |
224 | ||
225 | break; | |
226 | } | |
227 | ||
912d35f8 | 228 | if (spd->flags & SPLICE_F_NONBLOCK) { |
29e35094 LT |
229 | if (!ret) |
230 | ret = -EAGAIN; | |
231 | break; | |
232 | } | |
233 | ||
5274f052 JA |
234 | if (signal_pending(current)) { |
235 | if (!ret) | |
236 | ret = -ERESTARTSYS; | |
237 | break; | |
238 | } | |
239 | ||
240 | if (do_wakeup) { | |
c0bd1f65 | 241 | smp_mb(); |
3a326a2c IM |
242 | if (waitqueue_active(&pipe->wait)) |
243 | wake_up_interruptible_sync(&pipe->wait); | |
244 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); | |
5274f052 JA |
245 | do_wakeup = 0; |
246 | } | |
247 | ||
3a326a2c IM |
248 | pipe->waiting_writers++; |
249 | pipe_wait(pipe); | |
250 | pipe->waiting_writers--; | |
5274f052 JA |
251 | } |
252 | ||
3a326a2c IM |
253 | if (pipe->inode) |
254 | mutex_unlock(&pipe->inode->i_mutex); | |
5274f052 JA |
255 | |
256 | if (do_wakeup) { | |
c0bd1f65 | 257 | smp_mb(); |
3a326a2c IM |
258 | if (waitqueue_active(&pipe->wait)) |
259 | wake_up_interruptible(&pipe->wait); | |
260 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); | |
5274f052 JA |
261 | } |
262 | ||
912d35f8 JA |
263 | while (page_nr < spd->nr_pages) |
264 | page_cache_release(spd->pages[page_nr++]); | |
5274f052 JA |
265 | |
266 | return ret; | |
267 | } | |
268 | ||
3a326a2c | 269 | static int |
cbb7e577 JA |
270 | __generic_file_splice_read(struct file *in, loff_t *ppos, |
271 | struct pipe_inode_info *pipe, size_t len, | |
272 | unsigned int flags) | |
5274f052 JA |
273 | { |
274 | struct address_space *mapping = in->f_mapping; | |
912d35f8 | 275 | unsigned int loff, nr_pages; |
16c523dd | 276 | struct page *pages[PIPE_BUFFERS]; |
912d35f8 | 277 | struct partial_page partial[PIPE_BUFFERS]; |
5274f052 | 278 | struct page *page; |
91ad66ef JA |
279 | pgoff_t index, end_index; |
280 | loff_t isize; | |
912d35f8 | 281 | size_t total_len; |
eb20796b | 282 | int error, page_nr; |
912d35f8 JA |
283 | struct splice_pipe_desc spd = { |
284 | .pages = pages, | |
285 | .partial = partial, | |
286 | .flags = flags, | |
287 | .ops = &page_cache_pipe_buf_ops, | |
288 | }; | |
5274f052 | 289 | |
cbb7e577 | 290 | index = *ppos >> PAGE_CACHE_SHIFT; |
912d35f8 JA |
291 | loff = *ppos & ~PAGE_CACHE_MASK; |
292 | nr_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
5274f052 JA |
293 | |
294 | if (nr_pages > PIPE_BUFFERS) | |
295 | nr_pages = PIPE_BUFFERS; | |
296 | ||
297 | /* | |
73d62d83 | 298 | * Initiate read-ahead on this page range. however, don't call into |
0b749ce3 JA |
299 | * read-ahead if this is a non-zero offset (we are likely doing small |
300 | * chunk splice and the page is already there) for a single page. | |
5274f052 | 301 | */ |
eb645a24 JA |
302 | if (!loff || nr_pages > 1) |
303 | page_cache_readahead(mapping, &in->f_ra, in, index, nr_pages); | |
5274f052 | 304 | |
5274f052 | 305 | /* |
73d62d83 | 306 | * Now fill in the holes: |
5274f052 | 307 | */ |
7480a904 | 308 | error = 0; |
912d35f8 | 309 | total_len = 0; |
82aa5d61 | 310 | |
eb20796b JA |
311 | /* |
312 | * Lookup the (hopefully) full range of pages we need. | |
313 | */ | |
314 | spd.nr_pages = find_get_pages_contig(mapping, index, nr_pages, pages); | |
82aa5d61 | 315 | |
eb20796b JA |
316 | /* |
317 | * If find_get_pages_contig() returned fewer pages than we needed, | |
318 | * allocate the rest. | |
319 | */ | |
320 | index += spd.nr_pages; | |
321 | while (spd.nr_pages < nr_pages) { | |
82aa5d61 | 322 | /* |
eb20796b JA |
323 | * Page could be there, find_get_pages_contig() breaks on |
324 | * the first hole. | |
5274f052 | 325 | */ |
7480a904 JA |
326 | page = find_get_page(mapping, index); |
327 | if (!page) { | |
7480a904 | 328 | /* |
eb20796b | 329 | * page didn't exist, allocate one. |
7480a904 JA |
330 | */ |
331 | page = page_cache_alloc_cold(mapping); | |
332 | if (!page) | |
333 | break; | |
334 | ||
335 | error = add_to_page_cache_lru(page, mapping, index, | |
eb20796b | 336 | mapping_gfp_mask(mapping)); |
7480a904 JA |
337 | if (unlikely(error)) { |
338 | page_cache_release(page); | |
339 | break; | |
340 | } | |
eb20796b JA |
341 | /* |
342 | * add_to_page_cache() locks the page, unlock it | |
343 | * to avoid convoluting the logic below even more. | |
344 | */ | |
345 | unlock_page(page); | |
7480a904 JA |
346 | } |
347 | ||
eb20796b JA |
348 | pages[spd.nr_pages++] = page; |
349 | index++; | |
350 | } | |
351 | ||
352 | /* | |
353 | * Now loop over the map and see if we need to start IO on any | |
354 | * pages, fill in the partial map, etc. | |
355 | */ | |
356 | index = *ppos >> PAGE_CACHE_SHIFT; | |
357 | nr_pages = spd.nr_pages; | |
358 | spd.nr_pages = 0; | |
359 | for (page_nr = 0; page_nr < nr_pages; page_nr++) { | |
360 | unsigned int this_len; | |
361 | ||
362 | if (!len) | |
363 | break; | |
364 | ||
365 | /* | |
366 | * this_len is the max we'll use from this page | |
367 | */ | |
368 | this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff); | |
369 | page = pages[page_nr]; | |
370 | ||
7480a904 JA |
371 | /* |
372 | * If the page isn't uptodate, we may need to start io on it | |
373 | */ | |
374 | if (!PageUptodate(page)) { | |
c4f895cb JA |
375 | /* |
376 | * If in nonblock mode then dont block on waiting | |
377 | * for an in-flight io page | |
378 | */ | |
379 | if (flags & SPLICE_F_NONBLOCK) | |
380 | break; | |
381 | ||
7480a904 JA |
382 | lock_page(page); |
383 | ||
384 | /* | |
385 | * page was truncated, stop here. if this isn't the | |
386 | * first page, we'll just complete what we already | |
387 | * added | |
388 | */ | |
389 | if (!page->mapping) { | |
390 | unlock_page(page); | |
7480a904 JA |
391 | break; |
392 | } | |
393 | /* | |
394 | * page was already under io and is now done, great | |
395 | */ | |
396 | if (PageUptodate(page)) { | |
397 | unlock_page(page); | |
398 | goto fill_it; | |
399 | } | |
5274f052 | 400 | |
7480a904 JA |
401 | /* |
402 | * need to read in the page | |
403 | */ | |
404 | error = mapping->a_ops->readpage(in, page); | |
5274f052 | 405 | if (unlikely(error)) { |
eb20796b JA |
406 | /* |
407 | * We really should re-lookup the page here, | |
408 | * but it complicates things a lot. Instead | |
409 | * lets just do what we already stored, and | |
410 | * we'll get it the next time we are called. | |
411 | */ | |
7480a904 | 412 | if (error == AOP_TRUNCATED_PAGE) |
eb20796b JA |
413 | error = 0; |
414 | ||
5274f052 JA |
415 | break; |
416 | } | |
91ad66ef JA |
417 | |
418 | /* | |
419 | * i_size must be checked after ->readpage(). | |
420 | */ | |
421 | isize = i_size_read(mapping->host); | |
422 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; | |
eb20796b | 423 | if (unlikely(!isize || index > end_index)) |
91ad66ef | 424 | break; |
91ad66ef JA |
425 | |
426 | /* | |
427 | * if this is the last page, see if we need to shrink | |
428 | * the length and stop | |
429 | */ | |
430 | if (end_index == index) { | |
431 | loff = PAGE_CACHE_SIZE - (isize & ~PAGE_CACHE_MASK); | |
eb20796b | 432 | if (total_len + loff > isize) |
91ad66ef | 433 | break; |
91ad66ef JA |
434 | /* |
435 | * force quit after adding this page | |
436 | */ | |
eb20796b | 437 | len = this_len; |
82aa5d61 | 438 | this_len = min(this_len, loff); |
912d35f8 | 439 | loff = 0; |
91ad66ef | 440 | } |
5274f052 | 441 | } |
7480a904 | 442 | fill_it: |
eb20796b JA |
443 | partial[page_nr].offset = loff; |
444 | partial[page_nr].len = this_len; | |
82aa5d61 | 445 | len -= this_len; |
912d35f8 | 446 | total_len += this_len; |
91ad66ef | 447 | loff = 0; |
eb20796b JA |
448 | spd.nr_pages++; |
449 | index++; | |
5274f052 JA |
450 | } |
451 | ||
eb20796b JA |
452 | /* |
453 | * Release any pages at the end, if we quit early. 'i' is how far | |
454 | * we got, 'nr_pages' is how many pages are in the map. | |
455 | */ | |
456 | while (page_nr < nr_pages) | |
457 | page_cache_release(pages[page_nr++]); | |
458 | ||
912d35f8 | 459 | if (spd.nr_pages) |
00522fb4 | 460 | return splice_to_pipe(pipe, &spd); |
5274f052 | 461 | |
7480a904 | 462 | return error; |
5274f052 JA |
463 | } |
464 | ||
83f9135b JA |
465 | /** |
466 | * generic_file_splice_read - splice data from file to a pipe | |
467 | * @in: file to splice from | |
468 | * @pipe: pipe to splice to | |
469 | * @len: number of bytes to splice | |
470 | * @flags: splice modifier flags | |
471 | * | |
472 | * Will read pages from given file and fill them into a pipe. | |
83f9135b | 473 | */ |
cbb7e577 JA |
474 | ssize_t generic_file_splice_read(struct file *in, loff_t *ppos, |
475 | struct pipe_inode_info *pipe, size_t len, | |
476 | unsigned int flags) | |
5274f052 JA |
477 | { |
478 | ssize_t spliced; | |
479 | int ret; | |
480 | ||
481 | ret = 0; | |
482 | spliced = 0; | |
3a326a2c | 483 | |
5274f052 | 484 | while (len) { |
cbb7e577 | 485 | ret = __generic_file_splice_read(in, ppos, pipe, len, flags); |
5274f052 | 486 | |
c4f895cb | 487 | if (ret < 0) |
5274f052 | 488 | break; |
c4f895cb JA |
489 | else if (!ret) { |
490 | if (spliced) | |
491 | break; | |
492 | if (flags & SPLICE_F_NONBLOCK) { | |
493 | ret = -EAGAIN; | |
494 | break; | |
495 | } | |
496 | } | |
5274f052 | 497 | |
cbb7e577 | 498 | *ppos += ret; |
5274f052 JA |
499 | len -= ret; |
500 | spliced += ret; | |
501 | } | |
502 | ||
503 | if (spliced) | |
504 | return spliced; | |
505 | ||
506 | return ret; | |
507 | } | |
508 | ||
059a8f37 JA |
509 | EXPORT_SYMBOL(generic_file_splice_read); |
510 | ||
5274f052 | 511 | /* |
4f6f0bd2 | 512 | * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos' |
016b661e | 513 | * using sendpage(). Return the number of bytes sent. |
5274f052 JA |
514 | */ |
515 | static int pipe_to_sendpage(struct pipe_inode_info *info, | |
516 | struct pipe_buffer *buf, struct splice_desc *sd) | |
517 | { | |
518 | struct file *file = sd->file; | |
519 | loff_t pos = sd->pos; | |
5274f052 JA |
520 | ssize_t ret; |
521 | void *ptr; | |
b2b39fa4 | 522 | int more; |
5274f052 JA |
523 | |
524 | /* | |
73d62d83 | 525 | * Sub-optimal, but we are limited by the pipe ->map. We don't |
5274f052 JA |
526 | * need a kmap'ed buffer here, we just want to make sure we |
527 | * have the page pinned if the pipe page originates from the | |
73d62d83 | 528 | * page cache. |
5274f052 JA |
529 | */ |
530 | ptr = buf->ops->map(file, info, buf); | |
531 | if (IS_ERR(ptr)) | |
532 | return PTR_ERR(ptr); | |
533 | ||
b2b39fa4 | 534 | more = (sd->flags & SPLICE_F_MORE) || sd->len < sd->total_len; |
5274f052 | 535 | |
016b661e JA |
536 | ret = file->f_op->sendpage(file, buf->page, buf->offset, sd->len, |
537 | &pos, more); | |
5274f052 JA |
538 | |
539 | buf->ops->unmap(info, buf); | |
016b661e | 540 | return ret; |
5274f052 JA |
541 | } |
542 | ||
543 | /* | |
544 | * This is a little more tricky than the file -> pipe splicing. There are | |
545 | * basically three cases: | |
546 | * | |
547 | * - Destination page already exists in the address space and there | |
548 | * are users of it. For that case we have no other option that | |
549 | * copying the data. Tough luck. | |
550 | * - Destination page already exists in the address space, but there | |
551 | * are no users of it. Make sure it's uptodate, then drop it. Fall | |
552 | * through to last case. | |
553 | * - Destination page does not exist, we can add the pipe page to | |
554 | * the page cache and avoid the copy. | |
555 | * | |
83f9135b JA |
556 | * If asked to move pages to the output file (SPLICE_F_MOVE is set in |
557 | * sd->flags), we attempt to migrate pages from the pipe to the output | |
558 | * file address space page cache. This is possible if no one else has | |
559 | * the pipe page referenced outside of the pipe and page cache. If | |
560 | * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create | |
561 | * a new page in the output file page cache and fill/dirty that. | |
5274f052 JA |
562 | */ |
563 | static int pipe_to_file(struct pipe_inode_info *info, struct pipe_buffer *buf, | |
564 | struct splice_desc *sd) | |
565 | { | |
566 | struct file *file = sd->file; | |
567 | struct address_space *mapping = file->f_mapping; | |
3e7ee3e7 | 568 | gfp_t gfp_mask = mapping_gfp_mask(mapping); |
016b661e | 569 | unsigned int offset, this_len; |
5274f052 | 570 | struct page *page; |
5274f052 | 571 | pgoff_t index; |
5abc97aa | 572 | char *src; |
3e7ee3e7 | 573 | int ret; |
5274f052 JA |
574 | |
575 | /* | |
49d0b21b | 576 | * make sure the data in this buffer is uptodate |
5274f052 JA |
577 | */ |
578 | src = buf->ops->map(file, info, buf); | |
579 | if (IS_ERR(src)) | |
580 | return PTR_ERR(src); | |
581 | ||
582 | index = sd->pos >> PAGE_CACHE_SHIFT; | |
583 | offset = sd->pos & ~PAGE_CACHE_MASK; | |
584 | ||
016b661e JA |
585 | this_len = sd->len; |
586 | if (this_len + offset > PAGE_CACHE_SIZE) | |
587 | this_len = PAGE_CACHE_SIZE - offset; | |
588 | ||
5274f052 | 589 | /* |
0568b409 JA |
590 | * Reuse buf page, if SPLICE_F_MOVE is set and we are doing a full |
591 | * page. | |
5274f052 | 592 | */ |
0568b409 | 593 | if ((sd->flags & SPLICE_F_MOVE) && this_len == PAGE_CACHE_SIZE) { |
83f9135b JA |
594 | /* |
595 | * If steal succeeds, buf->page is now pruned from the vm | |
9e0267c2 JA |
596 | * side (LRU and page cache) and we can reuse it. The page |
597 | * will also be looked on successful return. | |
83f9135b | 598 | */ |
5abc97aa JA |
599 | if (buf->ops->steal(info, buf)) |
600 | goto find_page; | |
601 | ||
602 | page = buf->page; | |
46e678c9 JA |
603 | if (add_to_page_cache(page, mapping, index, gfp_mask)) { |
604 | unlock_page(page); | |
5abc97aa | 605 | goto find_page; |
46e678c9 | 606 | } |
3e7ee3e7 | 607 | |
0568b409 JA |
608 | page_cache_get(page); |
609 | ||
3e7ee3e7 JA |
610 | if (!(buf->flags & PIPE_BUF_FLAG_LRU)) |
611 | lru_cache_add(page); | |
5abc97aa JA |
612 | } else { |
613 | find_page: | |
9e0267c2 JA |
614 | page = find_lock_page(mapping, index); |
615 | if (!page) { | |
616 | ret = -ENOMEM; | |
617 | page = page_cache_alloc_cold(mapping); | |
618 | if (unlikely(!page)) | |
619 | goto out_nomem; | |
620 | ||
621 | /* | |
622 | * This will also lock the page | |
623 | */ | |
624 | ret = add_to_page_cache_lru(page, mapping, index, | |
625 | gfp_mask); | |
626 | if (unlikely(ret)) | |
627 | goto out; | |
628 | } | |
5abc97aa JA |
629 | |
630 | /* | |
9e0267c2 JA |
631 | * We get here with the page locked. If the page is also |
632 | * uptodate, we don't need to do more. If it isn't, we | |
633 | * may need to bring it in if we are not going to overwrite | |
634 | * the full page. | |
5abc97aa JA |
635 | */ |
636 | if (!PageUptodate(page)) { | |
016b661e | 637 | if (this_len < PAGE_CACHE_SIZE) { |
5abc97aa JA |
638 | ret = mapping->a_ops->readpage(file, page); |
639 | if (unlikely(ret)) | |
640 | goto out; | |
641 | ||
642 | lock_page(page); | |
643 | ||
644 | if (!PageUptodate(page)) { | |
645 | /* | |
73d62d83 | 646 | * Page got invalidated, repeat. |
5abc97aa JA |
647 | */ |
648 | if (!page->mapping) { | |
649 | unlock_page(page); | |
650 | page_cache_release(page); | |
651 | goto find_page; | |
652 | } | |
653 | ret = -EIO; | |
654 | goto out; | |
5274f052 | 655 | } |
9e0267c2 | 656 | } else |
5abc97aa | 657 | SetPageUptodate(page); |
5274f052 JA |
658 | } |
659 | } | |
660 | ||
016b661e | 661 | ret = mapping->a_ops->prepare_write(file, page, offset, offset+this_len); |
4f6f0bd2 JA |
662 | if (ret == AOP_TRUNCATED_PAGE) { |
663 | page_cache_release(page); | |
664 | goto find_page; | |
665 | } else if (ret) | |
5274f052 JA |
666 | goto out; |
667 | ||
0568b409 | 668 | if (buf->page != page) { |
5abc97aa JA |
669 | char *dst = kmap_atomic(page, KM_USER0); |
670 | ||
016b661e | 671 | memcpy(dst + offset, src + buf->offset, this_len); |
5abc97aa JA |
672 | flush_dcache_page(page); |
673 | kunmap_atomic(dst, KM_USER0); | |
674 | } | |
5274f052 | 675 | |
016b661e | 676 | ret = mapping->a_ops->commit_write(file, page, offset, offset+this_len); |
0568b409 JA |
677 | if (!ret) { |
678 | /* | |
679 | * Return the number of bytes written and mark page as | |
680 | * accessed, we are now done! | |
681 | */ | |
682 | ret = this_len; | |
683 | mark_page_accessed(page); | |
684 | balance_dirty_pages_ratelimited(mapping); | |
685 | } else if (ret == AOP_TRUNCATED_PAGE) { | |
4f6f0bd2 JA |
686 | page_cache_release(page); |
687 | goto find_page; | |
0568b409 | 688 | } |
5274f052 | 689 | out: |
0568b409 | 690 | page_cache_release(page); |
9e0267c2 | 691 | unlock_page(page); |
9aefe431 | 692 | out_nomem: |
5274f052 JA |
693 | buf->ops->unmap(info, buf); |
694 | return ret; | |
695 | } | |
696 | ||
83f9135b JA |
697 | /* |
698 | * Pipe input worker. Most of this logic works like a regular pipe, the | |
699 | * key here is the 'actor' worker passed in that actually moves the data | |
700 | * to the wanted destination. See pipe_to_file/pipe_to_sendpage above. | |
701 | */ | |
00522fb4 JA |
702 | ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out, |
703 | loff_t *ppos, size_t len, unsigned int flags, | |
704 | splice_actor *actor) | |
5274f052 | 705 | { |
5274f052 JA |
706 | int ret, do_wakeup, err; |
707 | struct splice_desc sd; | |
708 | ||
709 | ret = 0; | |
710 | do_wakeup = 0; | |
711 | ||
712 | sd.total_len = len; | |
713 | sd.flags = flags; | |
714 | sd.file = out; | |
cbb7e577 | 715 | sd.pos = *ppos; |
5274f052 | 716 | |
3a326a2c IM |
717 | if (pipe->inode) |
718 | mutex_lock(&pipe->inode->i_mutex); | |
5274f052 | 719 | |
5274f052 | 720 | for (;;) { |
6f767b04 JA |
721 | if (pipe->nrbufs) { |
722 | struct pipe_buffer *buf = pipe->bufs + pipe->curbuf; | |
5274f052 JA |
723 | struct pipe_buf_operations *ops = buf->ops; |
724 | ||
725 | sd.len = buf->len; | |
726 | if (sd.len > sd.total_len) | |
727 | sd.len = sd.total_len; | |
728 | ||
3a326a2c | 729 | err = actor(pipe, buf, &sd); |
016b661e | 730 | if (err <= 0) { |
5274f052 JA |
731 | if (!ret && err != -ENODATA) |
732 | ret = err; | |
733 | ||
734 | break; | |
735 | } | |
736 | ||
016b661e JA |
737 | ret += err; |
738 | buf->offset += err; | |
739 | buf->len -= err; | |
740 | ||
741 | sd.len -= err; | |
742 | sd.pos += err; | |
743 | sd.total_len -= err; | |
744 | if (sd.len) | |
745 | continue; | |
73d62d83 | 746 | |
5274f052 JA |
747 | if (!buf->len) { |
748 | buf->ops = NULL; | |
3a326a2c | 749 | ops->release(pipe, buf); |
6f767b04 JA |
750 | pipe->curbuf = (pipe->curbuf + 1) & (PIPE_BUFFERS - 1); |
751 | pipe->nrbufs--; | |
752 | if (pipe->inode) | |
753 | do_wakeup = 1; | |
5274f052 JA |
754 | } |
755 | ||
5274f052 JA |
756 | if (!sd.total_len) |
757 | break; | |
758 | } | |
759 | ||
6f767b04 | 760 | if (pipe->nrbufs) |
5274f052 | 761 | continue; |
3a326a2c | 762 | if (!pipe->writers) |
5274f052 | 763 | break; |
3a326a2c | 764 | if (!pipe->waiting_writers) { |
5274f052 JA |
765 | if (ret) |
766 | break; | |
767 | } | |
768 | ||
29e35094 LT |
769 | if (flags & SPLICE_F_NONBLOCK) { |
770 | if (!ret) | |
771 | ret = -EAGAIN; | |
772 | break; | |
773 | } | |
774 | ||
5274f052 JA |
775 | if (signal_pending(current)) { |
776 | if (!ret) | |
777 | ret = -ERESTARTSYS; | |
778 | break; | |
779 | } | |
780 | ||
781 | if (do_wakeup) { | |
c0bd1f65 | 782 | smp_mb(); |
3a326a2c IM |
783 | if (waitqueue_active(&pipe->wait)) |
784 | wake_up_interruptible_sync(&pipe->wait); | |
785 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); | |
5274f052 JA |
786 | do_wakeup = 0; |
787 | } | |
788 | ||
3a326a2c | 789 | pipe_wait(pipe); |
5274f052 JA |
790 | } |
791 | ||
3a326a2c IM |
792 | if (pipe->inode) |
793 | mutex_unlock(&pipe->inode->i_mutex); | |
5274f052 JA |
794 | |
795 | if (do_wakeup) { | |
c0bd1f65 | 796 | smp_mb(); |
3a326a2c IM |
797 | if (waitqueue_active(&pipe->wait)) |
798 | wake_up_interruptible(&pipe->wait); | |
799 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); | |
5274f052 JA |
800 | } |
801 | ||
5274f052 | 802 | return ret; |
5274f052 JA |
803 | } |
804 | ||
83f9135b JA |
805 | /** |
806 | * generic_file_splice_write - splice data from a pipe to a file | |
3a326a2c | 807 | * @pipe: pipe info |
83f9135b JA |
808 | * @out: file to write to |
809 | * @len: number of bytes to splice | |
810 | * @flags: splice modifier flags | |
811 | * | |
812 | * Will either move or copy pages (determined by @flags options) from | |
813 | * the given pipe inode to the given file. | |
814 | * | |
815 | */ | |
3a326a2c IM |
816 | ssize_t |
817 | generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out, | |
cbb7e577 | 818 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 819 | { |
4f6f0bd2 | 820 | struct address_space *mapping = out->f_mapping; |
3a326a2c IM |
821 | ssize_t ret; |
822 | ||
00522fb4 | 823 | ret = splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_file); |
a4514ebd | 824 | if (ret > 0) { |
4f6f0bd2 | 825 | struct inode *inode = mapping->host; |
4f6f0bd2 | 826 | |
a4514ebd JA |
827 | *ppos += ret; |
828 | ||
829 | /* | |
830 | * If file or inode is SYNC and we actually wrote some data, | |
831 | * sync it. | |
832 | */ | |
833 | if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) { | |
834 | int err; | |
835 | ||
836 | mutex_lock(&inode->i_mutex); | |
837 | err = generic_osync_inode(inode, mapping, | |
838 | OSYNC_METADATA|OSYNC_DATA); | |
839 | mutex_unlock(&inode->i_mutex); | |
4f6f0bd2 | 840 | |
a4514ebd JA |
841 | if (err) |
842 | ret = err; | |
843 | } | |
4f6f0bd2 JA |
844 | } |
845 | ||
846 | return ret; | |
5274f052 JA |
847 | } |
848 | ||
059a8f37 JA |
849 | EXPORT_SYMBOL(generic_file_splice_write); |
850 | ||
83f9135b JA |
851 | /** |
852 | * generic_splice_sendpage - splice data from a pipe to a socket | |
853 | * @inode: pipe inode | |
854 | * @out: socket to write to | |
855 | * @len: number of bytes to splice | |
856 | * @flags: splice modifier flags | |
857 | * | |
858 | * Will send @len bytes from the pipe to a network socket. No data copying | |
859 | * is involved. | |
860 | * | |
861 | */ | |
3a326a2c | 862 | ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out, |
cbb7e577 | 863 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 864 | { |
00522fb4 | 865 | return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage); |
5274f052 JA |
866 | } |
867 | ||
059a8f37 | 868 | EXPORT_SYMBOL(generic_splice_sendpage); |
a0f06780 | 869 | |
83f9135b JA |
870 | /* |
871 | * Attempt to initiate a splice from pipe to file. | |
872 | */ | |
3a326a2c | 873 | static long do_splice_from(struct pipe_inode_info *pipe, struct file *out, |
cbb7e577 | 874 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 875 | { |
5274f052 JA |
876 | int ret; |
877 | ||
49570e9b | 878 | if (unlikely(!out->f_op || !out->f_op->splice_write)) |
5274f052 JA |
879 | return -EINVAL; |
880 | ||
49570e9b | 881 | if (unlikely(!(out->f_mode & FMODE_WRITE))) |
5274f052 JA |
882 | return -EBADF; |
883 | ||
cbb7e577 | 884 | ret = rw_verify_area(WRITE, out, ppos, len); |
5274f052 JA |
885 | if (unlikely(ret < 0)) |
886 | return ret; | |
887 | ||
cbb7e577 | 888 | return out->f_op->splice_write(pipe, out, ppos, len, flags); |
5274f052 JA |
889 | } |
890 | ||
83f9135b JA |
891 | /* |
892 | * Attempt to initiate a splice from a file to a pipe. | |
893 | */ | |
cbb7e577 JA |
894 | static long do_splice_to(struct file *in, loff_t *ppos, |
895 | struct pipe_inode_info *pipe, size_t len, | |
896 | unsigned int flags) | |
5274f052 | 897 | { |
cbb7e577 | 898 | loff_t isize, left; |
5274f052 JA |
899 | int ret; |
900 | ||
49570e9b | 901 | if (unlikely(!in->f_op || !in->f_op->splice_read)) |
5274f052 JA |
902 | return -EINVAL; |
903 | ||
49570e9b | 904 | if (unlikely(!(in->f_mode & FMODE_READ))) |
5274f052 JA |
905 | return -EBADF; |
906 | ||
cbb7e577 | 907 | ret = rw_verify_area(READ, in, ppos, len); |
5274f052 JA |
908 | if (unlikely(ret < 0)) |
909 | return ret; | |
910 | ||
911 | isize = i_size_read(in->f_mapping->host); | |
cbb7e577 | 912 | if (unlikely(*ppos >= isize)) |
5274f052 JA |
913 | return 0; |
914 | ||
cbb7e577 | 915 | left = isize - *ppos; |
49570e9b | 916 | if (unlikely(left < len)) |
5274f052 JA |
917 | len = left; |
918 | ||
cbb7e577 | 919 | return in->f_op->splice_read(in, ppos, pipe, len, flags); |
5274f052 JA |
920 | } |
921 | ||
cbb7e577 JA |
922 | long do_splice_direct(struct file *in, loff_t *ppos, struct file *out, |
923 | size_t len, unsigned int flags) | |
b92ce558 JA |
924 | { |
925 | struct pipe_inode_info *pipe; | |
926 | long ret, bytes; | |
cbb7e577 | 927 | loff_t out_off; |
b92ce558 JA |
928 | umode_t i_mode; |
929 | int i; | |
930 | ||
931 | /* | |
932 | * We require the input being a regular file, as we don't want to | |
933 | * randomly drop data for eg socket -> socket splicing. Use the | |
934 | * piped splicing for that! | |
935 | */ | |
936 | i_mode = in->f_dentry->d_inode->i_mode; | |
937 | if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode))) | |
938 | return -EINVAL; | |
939 | ||
940 | /* | |
941 | * neither in nor out is a pipe, setup an internal pipe attached to | |
942 | * 'out' and transfer the wanted data from 'in' to 'out' through that | |
943 | */ | |
944 | pipe = current->splice_pipe; | |
49570e9b | 945 | if (unlikely(!pipe)) { |
b92ce558 JA |
946 | pipe = alloc_pipe_info(NULL); |
947 | if (!pipe) | |
948 | return -ENOMEM; | |
949 | ||
950 | /* | |
951 | * We don't have an immediate reader, but we'll read the stuff | |
00522fb4 | 952 | * out of the pipe right after the splice_to_pipe(). So set |
b92ce558 JA |
953 | * PIPE_READERS appropriately. |
954 | */ | |
955 | pipe->readers = 1; | |
956 | ||
957 | current->splice_pipe = pipe; | |
958 | } | |
959 | ||
960 | /* | |
73d62d83 | 961 | * Do the splice. |
b92ce558 JA |
962 | */ |
963 | ret = 0; | |
964 | bytes = 0; | |
cbb7e577 | 965 | out_off = 0; |
b92ce558 JA |
966 | |
967 | while (len) { | |
968 | size_t read_len, max_read_len; | |
969 | ||
970 | /* | |
971 | * Do at most PIPE_BUFFERS pages worth of transfer: | |
972 | */ | |
973 | max_read_len = min(len, (size_t)(PIPE_BUFFERS*PAGE_SIZE)); | |
974 | ||
cbb7e577 | 975 | ret = do_splice_to(in, ppos, pipe, max_read_len, flags); |
b92ce558 JA |
976 | if (unlikely(ret < 0)) |
977 | goto out_release; | |
978 | ||
979 | read_len = ret; | |
980 | ||
981 | /* | |
982 | * NOTE: nonblocking mode only applies to the input. We | |
983 | * must not do the output in nonblocking mode as then we | |
984 | * could get stuck data in the internal pipe: | |
985 | */ | |
cbb7e577 | 986 | ret = do_splice_from(pipe, out, &out_off, read_len, |
b92ce558 JA |
987 | flags & ~SPLICE_F_NONBLOCK); |
988 | if (unlikely(ret < 0)) | |
989 | goto out_release; | |
990 | ||
991 | bytes += ret; | |
992 | len -= ret; | |
993 | ||
994 | /* | |
995 | * In nonblocking mode, if we got back a short read then | |
996 | * that was due to either an IO error or due to the | |
997 | * pagecache entry not being there. In the IO error case | |
998 | * the _next_ splice attempt will produce a clean IO error | |
999 | * return value (not a short read), so in both cases it's | |
1000 | * correct to break out of the loop here: | |
1001 | */ | |
1002 | if ((flags & SPLICE_F_NONBLOCK) && (read_len < max_read_len)) | |
1003 | break; | |
1004 | } | |
1005 | ||
1006 | pipe->nrbufs = pipe->curbuf = 0; | |
1007 | ||
1008 | return bytes; | |
1009 | ||
1010 | out_release: | |
1011 | /* | |
1012 | * If we did an incomplete transfer we must release | |
1013 | * the pipe buffers in question: | |
1014 | */ | |
1015 | for (i = 0; i < PIPE_BUFFERS; i++) { | |
1016 | struct pipe_buffer *buf = pipe->bufs + i; | |
1017 | ||
1018 | if (buf->ops) { | |
1019 | buf->ops->release(pipe, buf); | |
1020 | buf->ops = NULL; | |
1021 | } | |
1022 | } | |
1023 | pipe->nrbufs = pipe->curbuf = 0; | |
1024 | ||
1025 | /* | |
1026 | * If we transferred some data, return the number of bytes: | |
1027 | */ | |
1028 | if (bytes > 0) | |
1029 | return bytes; | |
1030 | ||
1031 | return ret; | |
1032 | } | |
1033 | ||
1034 | EXPORT_SYMBOL(do_splice_direct); | |
1035 | ||
83f9135b JA |
1036 | /* |
1037 | * Determine where to splice to/from. | |
1038 | */ | |
529565dc IM |
1039 | static long do_splice(struct file *in, loff_t __user *off_in, |
1040 | struct file *out, loff_t __user *off_out, | |
1041 | size_t len, unsigned int flags) | |
5274f052 | 1042 | { |
3a326a2c | 1043 | struct pipe_inode_info *pipe; |
cbb7e577 | 1044 | loff_t offset, *off; |
a4514ebd | 1045 | long ret; |
5274f052 | 1046 | |
3a326a2c | 1047 | pipe = in->f_dentry->d_inode->i_pipe; |
529565dc IM |
1048 | if (pipe) { |
1049 | if (off_in) | |
1050 | return -ESPIPE; | |
b92ce558 JA |
1051 | if (off_out) { |
1052 | if (out->f_op->llseek == no_llseek) | |
1053 | return -EINVAL; | |
cbb7e577 | 1054 | if (copy_from_user(&offset, off_out, sizeof(loff_t))) |
b92ce558 | 1055 | return -EFAULT; |
cbb7e577 JA |
1056 | off = &offset; |
1057 | } else | |
1058 | off = &out->f_pos; | |
529565dc | 1059 | |
a4514ebd JA |
1060 | ret = do_splice_from(pipe, out, off, len, flags); |
1061 | ||
1062 | if (off_out && copy_to_user(off_out, off, sizeof(loff_t))) | |
1063 | ret = -EFAULT; | |
1064 | ||
1065 | return ret; | |
529565dc | 1066 | } |
5274f052 | 1067 | |
3a326a2c | 1068 | pipe = out->f_dentry->d_inode->i_pipe; |
529565dc IM |
1069 | if (pipe) { |
1070 | if (off_out) | |
1071 | return -ESPIPE; | |
b92ce558 JA |
1072 | if (off_in) { |
1073 | if (in->f_op->llseek == no_llseek) | |
1074 | return -EINVAL; | |
cbb7e577 | 1075 | if (copy_from_user(&offset, off_in, sizeof(loff_t))) |
b92ce558 | 1076 | return -EFAULT; |
cbb7e577 JA |
1077 | off = &offset; |
1078 | } else | |
1079 | off = &in->f_pos; | |
529565dc | 1080 | |
a4514ebd JA |
1081 | ret = do_splice_to(in, off, pipe, len, flags); |
1082 | ||
1083 | if (off_in && copy_to_user(off_in, off, sizeof(loff_t))) | |
1084 | ret = -EFAULT; | |
1085 | ||
1086 | return ret; | |
529565dc | 1087 | } |
5274f052 JA |
1088 | |
1089 | return -EINVAL; | |
1090 | } | |
1091 | ||
912d35f8 JA |
1092 | /* |
1093 | * Map an iov into an array of pages and offset/length tupples. With the | |
1094 | * partial_page structure, we can map several non-contiguous ranges into | |
1095 | * our ones pages[] map instead of splitting that operation into pieces. | |
1096 | * Could easily be exported as a generic helper for other users, in which | |
1097 | * case one would probably want to add a 'max_nr_pages' parameter as well. | |
1098 | */ | |
1099 | static int get_iovec_page_array(const struct iovec __user *iov, | |
1100 | unsigned int nr_vecs, struct page **pages, | |
1101 | struct partial_page *partial) | |
1102 | { | |
1103 | int buffers = 0, error = 0; | |
1104 | ||
1105 | /* | |
1106 | * It's ok to take the mmap_sem for reading, even | |
1107 | * across a "get_user()". | |
1108 | */ | |
1109 | down_read(¤t->mm->mmap_sem); | |
1110 | ||
1111 | while (nr_vecs) { | |
1112 | unsigned long off, npages; | |
1113 | void __user *base; | |
1114 | size_t len; | |
1115 | int i; | |
1116 | ||
1117 | /* | |
1118 | * Get user address base and length for this iovec. | |
1119 | */ | |
1120 | error = get_user(base, &iov->iov_base); | |
1121 | if (unlikely(error)) | |
1122 | break; | |
1123 | error = get_user(len, &iov->iov_len); | |
1124 | if (unlikely(error)) | |
1125 | break; | |
1126 | ||
1127 | /* | |
1128 | * Sanity check this iovec. 0 read succeeds. | |
1129 | */ | |
1130 | if (unlikely(!len)) | |
1131 | break; | |
1132 | error = -EFAULT; | |
1133 | if (unlikely(!base)) | |
1134 | break; | |
1135 | ||
1136 | /* | |
1137 | * Get this base offset and number of pages, then map | |
1138 | * in the user pages. | |
1139 | */ | |
1140 | off = (unsigned long) base & ~PAGE_MASK; | |
1141 | npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1142 | if (npages > PIPE_BUFFERS - buffers) | |
1143 | npages = PIPE_BUFFERS - buffers; | |
1144 | ||
1145 | error = get_user_pages(current, current->mm, | |
1146 | (unsigned long) base, npages, 0, 0, | |
1147 | &pages[buffers], NULL); | |
1148 | ||
1149 | if (unlikely(error <= 0)) | |
1150 | break; | |
1151 | ||
1152 | /* | |
1153 | * Fill this contiguous range into the partial page map. | |
1154 | */ | |
1155 | for (i = 0; i < error; i++) { | |
1156 | const int plen = min_t(size_t, len, PAGE_SIZE) - off; | |
1157 | ||
1158 | partial[buffers].offset = off; | |
1159 | partial[buffers].len = plen; | |
1160 | ||
1161 | off = 0; | |
1162 | len -= plen; | |
1163 | buffers++; | |
1164 | } | |
1165 | ||
1166 | /* | |
1167 | * We didn't complete this iov, stop here since it probably | |
1168 | * means we have to move some of this into a pipe to | |
1169 | * be able to continue. | |
1170 | */ | |
1171 | if (len) | |
1172 | break; | |
1173 | ||
1174 | /* | |
1175 | * Don't continue if we mapped fewer pages than we asked for, | |
1176 | * or if we mapped the max number of pages that we have | |
1177 | * room for. | |
1178 | */ | |
1179 | if (error < npages || buffers == PIPE_BUFFERS) | |
1180 | break; | |
1181 | ||
1182 | nr_vecs--; | |
1183 | iov++; | |
1184 | } | |
1185 | ||
1186 | up_read(¤t->mm->mmap_sem); | |
1187 | ||
1188 | if (buffers) | |
1189 | return buffers; | |
1190 | ||
1191 | return error; | |
1192 | } | |
1193 | ||
1194 | /* | |
1195 | * vmsplice splices a user address range into a pipe. It can be thought of | |
1196 | * as splice-from-memory, where the regular splice is splice-from-file (or | |
1197 | * to file). In both cases the output is a pipe, naturally. | |
1198 | * | |
1199 | * Note that vmsplice only supports splicing _from_ user memory to a pipe, | |
1200 | * not the other way around. Splicing from user memory is a simple operation | |
1201 | * that can be supported without any funky alignment restrictions or nasty | |
1202 | * vm tricks. We simply map in the user memory and fill them into a pipe. | |
1203 | * The reverse isn't quite as easy, though. There are two possible solutions | |
1204 | * for that: | |
1205 | * | |
1206 | * - memcpy() the data internally, at which point we might as well just | |
1207 | * do a regular read() on the buffer anyway. | |
1208 | * - Lots of nasty vm tricks, that are neither fast nor flexible (it | |
1209 | * has restriction limitations on both ends of the pipe). | |
1210 | * | |
1211 | * Alas, it isn't here. | |
1212 | * | |
1213 | */ | |
1214 | static long do_vmsplice(struct file *file, const struct iovec __user *iov, | |
1215 | unsigned long nr_segs, unsigned int flags) | |
1216 | { | |
1217 | struct pipe_inode_info *pipe = file->f_dentry->d_inode->i_pipe; | |
1218 | struct page *pages[PIPE_BUFFERS]; | |
1219 | struct partial_page partial[PIPE_BUFFERS]; | |
1220 | struct splice_pipe_desc spd = { | |
1221 | .pages = pages, | |
1222 | .partial = partial, | |
1223 | .flags = flags, | |
1224 | .ops = &user_page_pipe_buf_ops, | |
1225 | }; | |
1226 | ||
1227 | if (unlikely(!pipe)) | |
1228 | return -EBADF; | |
1229 | if (unlikely(nr_segs > UIO_MAXIOV)) | |
1230 | return -EINVAL; | |
1231 | else if (unlikely(!nr_segs)) | |
1232 | return 0; | |
1233 | ||
1234 | spd.nr_pages = get_iovec_page_array(iov, nr_segs, pages, partial); | |
1235 | if (spd.nr_pages <= 0) | |
1236 | return spd.nr_pages; | |
1237 | ||
00522fb4 | 1238 | return splice_to_pipe(pipe, &spd); |
912d35f8 JA |
1239 | } |
1240 | ||
1241 | asmlinkage long sys_vmsplice(int fd, const struct iovec __user *iov, | |
1242 | unsigned long nr_segs, unsigned int flags) | |
1243 | { | |
1244 | struct file *file; | |
1245 | long error; | |
1246 | int fput; | |
1247 | ||
1248 | error = -EBADF; | |
1249 | file = fget_light(fd, &fput); | |
1250 | if (file) { | |
1251 | if (file->f_mode & FMODE_WRITE) | |
1252 | error = do_vmsplice(file, iov, nr_segs, flags); | |
1253 | ||
1254 | fput_light(file, fput); | |
1255 | } | |
1256 | ||
1257 | return error; | |
1258 | } | |
1259 | ||
529565dc IM |
1260 | asmlinkage long sys_splice(int fd_in, loff_t __user *off_in, |
1261 | int fd_out, loff_t __user *off_out, | |
1262 | size_t len, unsigned int flags) | |
5274f052 JA |
1263 | { |
1264 | long error; | |
1265 | struct file *in, *out; | |
1266 | int fput_in, fput_out; | |
1267 | ||
1268 | if (unlikely(!len)) | |
1269 | return 0; | |
1270 | ||
1271 | error = -EBADF; | |
529565dc | 1272 | in = fget_light(fd_in, &fput_in); |
5274f052 JA |
1273 | if (in) { |
1274 | if (in->f_mode & FMODE_READ) { | |
529565dc | 1275 | out = fget_light(fd_out, &fput_out); |
5274f052 JA |
1276 | if (out) { |
1277 | if (out->f_mode & FMODE_WRITE) | |
529565dc IM |
1278 | error = do_splice(in, off_in, |
1279 | out, off_out, | |
1280 | len, flags); | |
5274f052 JA |
1281 | fput_light(out, fput_out); |
1282 | } | |
1283 | } | |
1284 | ||
1285 | fput_light(in, fput_in); | |
1286 | } | |
1287 | ||
1288 | return error; | |
1289 | } | |
70524490 JA |
1290 | |
1291 | /* | |
1292 | * Link contents of ipipe to opipe. | |
1293 | */ | |
1294 | static int link_pipe(struct pipe_inode_info *ipipe, | |
1295 | struct pipe_inode_info *opipe, | |
1296 | size_t len, unsigned int flags) | |
1297 | { | |
1298 | struct pipe_buffer *ibuf, *obuf; | |
2a27250e JA |
1299 | int ret, do_wakeup, i, ipipe_first; |
1300 | ||
1301 | ret = do_wakeup = ipipe_first = 0; | |
70524490 JA |
1302 | |
1303 | /* | |
1304 | * Potential ABBA deadlock, work around it by ordering lock | |
1305 | * grabbing by inode address. Otherwise two different processes | |
1306 | * could deadlock (one doing tee from A -> B, the other from B -> A). | |
1307 | */ | |
1308 | if (ipipe->inode < opipe->inode) { | |
2a27250e | 1309 | ipipe_first = 1; |
70524490 JA |
1310 | mutex_lock(&ipipe->inode->i_mutex); |
1311 | mutex_lock(&opipe->inode->i_mutex); | |
1312 | } else { | |
1313 | mutex_lock(&opipe->inode->i_mutex); | |
1314 | mutex_lock(&ipipe->inode->i_mutex); | |
1315 | } | |
1316 | ||
1317 | for (i = 0;; i++) { | |
1318 | if (!opipe->readers) { | |
1319 | send_sig(SIGPIPE, current, 0); | |
1320 | if (!ret) | |
1321 | ret = -EPIPE; | |
1322 | break; | |
1323 | } | |
1324 | if (ipipe->nrbufs - i) { | |
1325 | ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (PIPE_BUFFERS - 1)); | |
1326 | ||
1327 | /* | |
1328 | * If we have room, fill this buffer | |
1329 | */ | |
1330 | if (opipe->nrbufs < PIPE_BUFFERS) { | |
1331 | int nbuf = (opipe->curbuf + opipe->nrbufs) & (PIPE_BUFFERS - 1); | |
1332 | ||
1333 | /* | |
1334 | * Get a reference to this pipe buffer, | |
1335 | * so we can copy the contents over. | |
1336 | */ | |
1337 | ibuf->ops->get(ipipe, ibuf); | |
1338 | ||
1339 | obuf = opipe->bufs + nbuf; | |
1340 | *obuf = *ibuf; | |
1341 | ||
1342 | if (obuf->len > len) | |
1343 | obuf->len = len; | |
1344 | ||
1345 | opipe->nrbufs++; | |
1346 | do_wakeup = 1; | |
1347 | ret += obuf->len; | |
1348 | len -= obuf->len; | |
1349 | ||
1350 | if (!len) | |
1351 | break; | |
1352 | if (opipe->nrbufs < PIPE_BUFFERS) | |
1353 | continue; | |
1354 | } | |
1355 | ||
1356 | /* | |
1357 | * We have input available, but no output room. | |
2a27250e JA |
1358 | * If we already copied data, return that. If we |
1359 | * need to drop the opipe lock, it must be ordered | |
1360 | * last to avoid deadlocks. | |
70524490 | 1361 | */ |
2a27250e | 1362 | if ((flags & SPLICE_F_NONBLOCK) || !ipipe_first) { |
70524490 JA |
1363 | if (!ret) |
1364 | ret = -EAGAIN; | |
1365 | break; | |
1366 | } | |
1367 | if (signal_pending(current)) { | |
1368 | if (!ret) | |
1369 | ret = -ERESTARTSYS; | |
1370 | break; | |
1371 | } | |
1372 | if (do_wakeup) { | |
1373 | smp_mb(); | |
1374 | if (waitqueue_active(&opipe->wait)) | |
1375 | wake_up_interruptible(&opipe->wait); | |
1376 | kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN); | |
1377 | do_wakeup = 0; | |
1378 | } | |
1379 | ||
1380 | opipe->waiting_writers++; | |
1381 | pipe_wait(opipe); | |
1382 | opipe->waiting_writers--; | |
1383 | continue; | |
1384 | } | |
1385 | ||
1386 | /* | |
1387 | * No input buffers, do the usual checks for available | |
1388 | * writers and blocking and wait if necessary | |
1389 | */ | |
1390 | if (!ipipe->writers) | |
1391 | break; | |
1392 | if (!ipipe->waiting_writers) { | |
1393 | if (ret) | |
1394 | break; | |
1395 | } | |
2a27250e JA |
1396 | /* |
1397 | * pipe_wait() drops the ipipe mutex. To avoid deadlocks | |
1398 | * with another process, we can only safely do that if | |
1399 | * the ipipe lock is ordered last. | |
1400 | */ | |
1401 | if ((flags & SPLICE_F_NONBLOCK) || ipipe_first) { | |
70524490 JA |
1402 | if (!ret) |
1403 | ret = -EAGAIN; | |
1404 | break; | |
1405 | } | |
1406 | if (signal_pending(current)) { | |
1407 | if (!ret) | |
1408 | ret = -ERESTARTSYS; | |
1409 | break; | |
1410 | } | |
1411 | ||
1412 | if (waitqueue_active(&ipipe->wait)) | |
1413 | wake_up_interruptible_sync(&ipipe->wait); | |
1414 | kill_fasync(&ipipe->fasync_writers, SIGIO, POLL_OUT); | |
1415 | ||
1416 | pipe_wait(ipipe); | |
1417 | } | |
1418 | ||
1419 | mutex_unlock(&ipipe->inode->i_mutex); | |
1420 | mutex_unlock(&opipe->inode->i_mutex); | |
1421 | ||
1422 | if (do_wakeup) { | |
1423 | smp_mb(); | |
1424 | if (waitqueue_active(&opipe->wait)) | |
1425 | wake_up_interruptible(&opipe->wait); | |
1426 | kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN); | |
1427 | } | |
1428 | ||
1429 | return ret; | |
1430 | } | |
1431 | ||
1432 | /* | |
1433 | * This is a tee(1) implementation that works on pipes. It doesn't copy | |
1434 | * any data, it simply references the 'in' pages on the 'out' pipe. | |
1435 | * The 'flags' used are the SPLICE_F_* variants, currently the only | |
1436 | * applicable one is SPLICE_F_NONBLOCK. | |
1437 | */ | |
1438 | static long do_tee(struct file *in, struct file *out, size_t len, | |
1439 | unsigned int flags) | |
1440 | { | |
1441 | struct pipe_inode_info *ipipe = in->f_dentry->d_inode->i_pipe; | |
1442 | struct pipe_inode_info *opipe = out->f_dentry->d_inode->i_pipe; | |
1443 | ||
1444 | /* | |
1445 | * Link ipipe to the two output pipes, consuming as we go along. | |
1446 | */ | |
1447 | if (ipipe && opipe) | |
1448 | return link_pipe(ipipe, opipe, len, flags); | |
1449 | ||
1450 | return -EINVAL; | |
1451 | } | |
1452 | ||
1453 | asmlinkage long sys_tee(int fdin, int fdout, size_t len, unsigned int flags) | |
1454 | { | |
1455 | struct file *in; | |
1456 | int error, fput_in; | |
1457 | ||
1458 | if (unlikely(!len)) | |
1459 | return 0; | |
1460 | ||
1461 | error = -EBADF; | |
1462 | in = fget_light(fdin, &fput_in); | |
1463 | if (in) { | |
1464 | if (in->f_mode & FMODE_READ) { | |
1465 | int fput_out; | |
1466 | struct file *out = fget_light(fdout, &fput_out); | |
1467 | ||
1468 | if (out) { | |
1469 | if (out->f_mode & FMODE_WRITE) | |
1470 | error = do_tee(in, out, len, flags); | |
1471 | fput_light(out, fput_out); | |
1472 | } | |
1473 | } | |
1474 | fput_light(in, fput_in); | |
1475 | } | |
1476 | ||
1477 | return error; | |
1478 | } |